aclnnExpSegsumBackward

Supported Products

Product	Supported
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	×
Atlas inference products	×
Atlas training products	×

Function

Description: Performs backward pass of aclnnExpSegsum.
Formula (using 5D input as an example):
1. Multiply the input gradOutput of shape (N1, N2, N3, N4, N4) with the input gradSelf (forward output).
  $out\_mul = gradOutput * gradSelf$
2. Generate a BOOL triangular matrix A of shape (N4, N4), where the upper triangle is True, the lower triangle is False, and the diagonal is True.
3. Fill the elements in $out\_mul$ with 0 at the positions where the corresponding elements in matrix A are True.
  $out\_mul_i= \begin{cases}out\_mul_i,\quad A_i==False \\0, \quad A_i==True \end{cases}$
4. Reverse the second-to-last dimension of out_mul to generate out_flip.
5. Perform a cumulative sum along the second-to-last dimension of $out\_flip$ . For an element from a dimensional perspective, other dimension indices remain unchanged while the current index increases. $out\_cumsum\_{i}$ is the element at the corresponding position in the output tensor.
  $out\_cumsum_{i} = out\_flip_{1} + out\_flip_{2} + out\_flip_{3} + ...... + out\_flip_{i}$
6. Reverse the second-to-last dimension of $out\_cumsum$ to generate out_flip2.
7. Generate a BOOL triangular matrix B of shape (N4, N4), where the upper triangle is True, the lower triangle is False, and the diagonal is True.
8. Fill the elements in $out\_flip2$ with 0 at the positions where the corresponding elements in matrix B are True.
  $out\_flip2_i= \begin{cases}out\_flip2_i,\quad B_i==False \\0, \quad B_i==True \end{cases}$
9. Return gradInput as the sum of each row along the last dimension of out\_flip2.

Prototype

Each operator has two-phase API calls. First, aclnnExpSegsumBackwardGetWorkspaceSize is called to obtain the workspace size required for computation and the executor that contains the operator computation process. Then, aclnnExpSegsumBackward is called to perform computation.

aclnnStatus aclnnExpSegsumBackwardGetWorkspaceSize(
  const aclTensor*     gradOutput,
  const aclTensor*     gradSelf,
  aclTensor           *gradInput,
  uint64_t            *workspaceSize,
  aclOpExecutor      **executor)

aclnnStatus aclnnExpSegsumBackward(
  void          *workspace,
  uint64_t       workspaceSize,
  aclOpExecutor *executor,
  aclrtStream    stream)

aclnnExpSegsumBackwardGetWorkspaceSize

Parameters:

Name	Input/Output	Description	Usage Notes	Data Type	Data Format	Dimension (Shape)	Non-contiguous Tensor
gradOutput	Input	Gradient for backward pass. It corresponds to gradOutput in the formula.	Empty tensors are supported. The last two dimensions of the shape must have the same size.	FLOAT32, FLOAT16, BFLOAT16	ND	4–5	√
gradSelf	Input	Forward pass output. It corresponds to gradSelf in the formula.	Empty tensors are supported. The shape and data type are the same as those of gradOutput.	FLOAT32, FLOAT16, BFLOAT16	ND	4–5	√
gradInput	Input	Backward pass output. It corresponds to gradInput in the formula.	Empty tensors are supported. The data type must be the same as that of gradOutput. The output dimension must be one less than the input dimension. 3D or 4D is supported. When the input gradOutput is 4D, the output dimension size is the same as the first three dimensions of gradOutput. When the input gradOutput is 5D, the output dimension size is the same as the first four dimensions of gradOutput.	FLOAT32, FLOAT16, BFLOAT16	ND	3–4	√
workspaceSize	Output	Size of the workspace required to be allocated on the device.	-	-	-	-	-
executor	Output	Operator executor, containing the operator computation process.	-	-	-	-	-

Returns:

aclnnStatus: status code. For details, see aclnn Return Codes.

The first-phase API performs input parameter validation. The following errors may be returned:

Return	Error Code	Description
ACLNN_ERR_PARAM_NULLPTR	161001	The passed gradOutput, gradSelf, or gradInput is a null pointer.
ACLNN_ERR_PARAM_INVALID	161002	The data type of gradOutput, gradSelf, or gradInput is not supported.
ACLNN_ERR_PARAM_INVALID	161002	The shape and data type of gradOutput, gradSelf, or gradInput do not meet the parameter requirements.

aclnnExpSegsumBackward

Parameters:

Name	Input/Output	Description
workspace	Input	Memory address of the workspace to be allocated on the device.
workspaceSize	Input	Size of the workspace to be allocated on the device, which is obtained by calling the first-phase API aclnnExpSegsumBackwardGetWorkspaceSize.
executor	Input	Operator executor, containing the operator computation process.
stream	Input	Stream for executing the task.

Returns:

aclnnStatus: status code. For details, see aclnn Return Codes.

Constraints

Deterministic computation:
- aclnnExpSegsumBackward defaults to a deterministic implementation.

Calling Example

The following example is for reference only. For details, see Compilation and Running Sample.

#include <iostream>
#include <vector>
#include "acl/acl.h"
#include "aclnnop/aclnn_exp_segsum_backward.h"

#define CHECK_RET(cond, return_expr) \
  do {                               \
    if (!(cond)) {                   \
      return_expr;                   \
    }                                \
  } while (0)

#define LOG_PRINT(message, ...)     \
  do {                              \
    printf(message, ##__VA_ARGS__); \
  } while (0)

int64_t GetShapeSize(const std::vector<int64_t>& shape) {
  int64_t shape_size = 1;
  for (auto i : shape) {
    shape_size *= i;
  }
  return shape_size;
}

int Init(int32_t deviceId, aclrtStream* stream) {
  // (Boilerplate) Initialize resources.
  auto ret = aclInit(nullptr);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret); return ret);
  ret = aclrtSetDevice(deviceId);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret);
  ret = aclrtCreateStream(stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret); return ret);
  return 0;
}

template <typename T>
int CreateAclTensor(const std::vector<T>& hostData, const std::vector<int64_t>& shape, void** deviceAddr,
                    aclDataType dataType, aclTensor** tensor) {
  auto size = GetShapeSize(shape) * sizeof(T);
  // Call aclrtMalloc to allocate device memory.
  auto ret = aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMalloc failed. ERROR: %d\n", ret); return ret);

  // Call aclrtMemcpy to copy host data to the device memory.
  ret = aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMemcpy failed. ERROR: %d\n", ret); return ret);

  // Compute the strides of the contiguous tensor.
  std::vector<int64_t> strides(shape.size(), 1);
  for (int64_t i = shape.size() - 2; i >= 0; i--) {
    strides[i] = shape[i + 1] * strides[i + 1];
  }

  // Call aclCreateTensor to create an aclTensor.
  *tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
                            shape.data(), shape.size(), *deviceAddr);
  return 0;
}

int main() {
  // 1. (Boilerplate) Initialize the device and stream. For details, see the ACL API manual.
  // Set the device ID based on the actual device.
  int32_t deviceId = 0;
  aclrtStream stream;
  auto ret = Init(deviceId, &stream);
  // Perform the check as required.
  CHECK_RET(ret == 0, LOG_PRINT("Init acl failed. ERROR: %d\n", ret); return ret);

  // 2. Construct inputs and outputs based on the API definition.
  void *gradOutputDeviceAddr = nullptr;
  void *gradSelfDeviceAddr = nullptr;
  void *gradInputDeviceAddr = nullptr;
  aclTensor *gradOutput = nullptr;
  aclTensor *gradSelf = nullptr;
  aclTensor *gradInput = nullptr;
  std::vector<int64_t> gradOutputShape = {1, 1, 1, 4, 4};
  std::vector<int64_t> gradSelfShape = {1, 1, 1, 4, 4};
  std::vector<int64_t> gradInputShape = {1, 1, 1, 4};
  std::vector<float> gradOutputHostData(16,1);
  std::vector<float> gradSelfHostData(16,1);
  std::vector<float> gradInputHostData(4, 0);

  // Create an input aclTensor.
  ret = CreateAclTensor(gradOutputHostData, gradOutputShape, &gradOutputDeviceAddr, aclDataType::ACL_FLOAT, &gradOutput);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  ret = CreateAclTensor(gradSelfHostData, gradSelfShape, &gradSelfDeviceAddr, aclDataType::ACL_FLOAT, &gradSelf);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // Create an out aclTensor.
  ret = CreateAclTensor(gradInputHostData, gradInputShape, &gradInputDeviceAddr, aclDataType::ACL_FLOAT, &gradInput);
  CHECK_RET(ret == ACL_SUCCESS, return ret);
  // 3. Call the CANN operator library API, which needs to be replaced with the actual API.
  uint64_t workspaceSize = 0;
  aclOpExecutor* executor;
  // Call the first-phase API of aclnnExpSegsumBackward.
  ret = aclnnExpSegsumBackwardGetWorkspaceSize(gradOutput, gradSelf, gradInput, &workspaceSize, &executor);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnExpSegsumBackwardGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
  // Allocate device memory based on workspaceSize computed by the first-phase API.
  void* workspaceAddr = nullptr;
  if (workspaceSize > 0) {
    ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
    CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret;);
  }
  // Call the second-phase API of aclnnExpSegsumBackward.
  ret = aclnnExpSegsumBackward(workspaceAddr, workspaceSize, executor, stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnExpSegsumBackward failed. ERROR: %d\n", ret); return ret);
  // 4. (Boilerplate) Wait until the task execution is complete.
  ret = aclrtSynchronizeStream(stream);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
  // 5. Obtain the output value and copy the result from the device memory to the host. Modify the code based on the API definition.
  auto size = GetShapeSize(gradInputShape);
  std::vector<float> resultData(size, 0);
  ret = aclrtMemcpy(resultData.data(), resultData.size() * sizeof(resultData[0]), gradInputDeviceAddr, size * sizeof(float),
                    ACL_MEMCPY_DEVICE_TO_HOST);
  CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("copy result from device to host failed. ERROR: %d\n", ret); return ret);
  for (int64_t i = 0; i < size; i++) {
    LOG_PRINT("result[%ld] is: %f\n", i, resultData[i]);
  }

  // 6. Destroy aclTensor. Modify the code based on the API definition.
  aclDestroyTensor(gradOutput);
  aclDestroyTensor(gradSelf);
  aclDestroyTensor(gradInput);
  
  // 7. Free device resources. Modify the code based on the API definition.
  aclrtFree(gradOutputDeviceAddr);
  aclrtFree(gradSelfDeviceAddr);
  aclrtFree(gradInputDeviceAddr);
  if (workspaceSize > 0) {
    aclrtFree(workspaceAddr);
  }
  aclrtDestroyStream(stream);
  aclrtResetDevice(deviceId);
  aclFinalize();
  return 0;
}